Meta-Learning Driven Graph Neural Network for Real-Time Dynamic Graph Trend Modeling and Prediction

Real-time trend prediction in dynamic social networks is a crucial yet challenging task due to the evolving nature of interactions and user behaviors. In this paper, we propose Meta-enhanced Graph Network (MeGNet), a novel framework that integrates gradient-based meta-learning methods such as MAML with dynamic graph neural networks (GNNs) to address these challenges. Specifically, MeGNet employs a dynamic GNN backbone to model temporal and structural evolutions within the network, combined with meta-learning to capture transferable knowledge across different social contexts for rapid adaptation to unseen dynamic subgraphs or events. Meanwhile, the graph neural backbone effectively models temporal and structural evolutions within the network. To further support real-time applications, MeGNet incorporates a lightweight adaptation mechanism that reduces computational overhead without sacrificing predictive accuracy. Experimental results on multiple real-world social network datasets demonstrate that MeGNet achieves superior performance in both trend prediction accuracy and adaptation speed, highlighting its potential for practical deployment in time-sensitive social media analysis tasks.